Jasdeep Singh Sahni

📌 Build Momentum Through Continuous Releases 📌   Releasing something every sprint keeps your team sharp and confident. Sharon Allpress, VP of Strategic Engagements at EOT.AI, shares how maintaining a continuous release rhythm prevents teams from getting rusty and strengthens collaboration across product, engineering, and stakeholders.   Here’s her approach: Release something every sprint, even small improvements Keep documentation and stakeholder comms updated Use momentum to stay ready for larger launches   🎥 Watch now on YouTube: https://lnkd.in/e-PgruY7

1

I just read “Enterprise AI’s Reality Check” by Kapil Venkatachalam at FTV Capital, and it's a breath of fresh air amidst the AI hype. Kapil cuts right to it: amid billions in infrastructure bets and LLM growth, the real opportunities lie deeper—at the application layer, where AI directly powers business value. From personalized assistants adding real revenue at companies like Toast, Taco Bell, and Bank of America, to startups like Orbus Software, Windward, Arden Insurance, and Kore.ai, AI is starting to deliver beyond pilot illusions.. If you're focused on bridging the gap between AI experimentation and business outcomes, this is a must-read! #aiforreal #AIAdoption #AIAdoption #AIApplications #DataGovernance #AIInfrastructure #AIAtScale https://lnkd.in/ggzyBexW

3

With AI adoption and deeper integration into enterprise products, the world moves toward being collectively intelligent. First movers will have a distinct advantage. Fully backward-integrated AI companies are best positioned to win on economics.

1

Launching: Chat with Priyank (beta) today on https://priyank.tech A few weeks ago, I launched priyank.tech as a single home for my work, projects, and notes. Today I’m adding a faster way to get to the signal: Skip the scrolling, ask.  “Chat with Priyank” answers questions about my projects, views, and skills by pulling the most relevant context and giving you a concise, useful response. What’s under the hood (and why): - Human‑verified access - Conversations start only after the system knows it’s interacting with a person, not a bot. - Scope & guardrails - It only answers questions about me. When a query is out of scope or uncertain, it responds with a polite “I don’t know.” - Context engineering first (no RAG, yet!) - An analysis by industry veterans suggested that GPT-5 class models can correctly use ~85% of the provided context. I want to see how far careful, prompt, and context design can go before adding retrieval. I’ll evaluate this over the next few days and decide before graduating from beta. - Prompt caching enabled - To keep costs under control (and responses snappy), repeat prompts and shared context are cached. - Early memory layer - A lightweight short-term memory layer powers more personal, consistent answers across sessions. Try it on: https://priyank.tech   If it’s your first visit, do subscribe to my newsletter—I’ll share build notes, decisions, and lessons learned there. Starter prompts: “Tell me about your AI and Generative AI expertise “How can you help me become an AI first organization?” Feedback is welcome, especially edge cases and failure modes.  #AI #product #buildinpublic #LLM

31

2 Comments

As we continue to help our customers and partners transform consumer experiences across industry sectors from healthcare to retail to media & entertainment, I am incredibly excited to share the announcement of Google Cloud’s strategic partnership with BRAHMA AI — a pioneering company scaling high-fidelity, interactive “digital humans” for global enterprises. Hear it from CEO Prabhu Narasimhan himself — or rather, from his digital human “ATMAN”.

21

1 Comment

“Human Edge in the AI Age” by Nitin Seth is a great read providing comprehensive strategies for humans to thrive and succeed in the AI World. Incedo Inc.

37

Amazon Web Services (AWS) just launched general availability of new R8gn instances - though the way that I found out about them wasn't from the official Amazon blog post, it was from the new automated newsletter for EC2instances.info. EC2instances.info is regularly rebuilding itself every few hours and now also checks for two things: 1) new instance types and 2) any pricing changes with existing instance types. You can sign up to receive email alerts about these two changes as often as the site finds them. If you're curious about getting the notifications sent your way to see what Amazon is launching, especially leading up to AWS re:Invent, I included the link for how to set this up in the comments below.

34

3 Comments

As more teams push LLMs toward the edge, understanding real-world deployment trade-offs is becoming mission-critical. What does it take to run LLMs on-device, without sacrificing speed, accuracy, or user experience? This deep dive from Sri Raghu Malireddi at Grammarly at MLOps World | GenAI Summit walks through practical, post-training optimization strategies for edge deployment. Packed with real benchmarks and architecture-level tips, it’s one of the clearest breakdowns I’ve seen on bringing LLMs to consumer hardware. 🔍 Key Learnings: • Why on-device inference matters: lower latency, better privacy, reduced infra cost • Optimization stack: quantization (BFloat16, FP16, INT4), pruning, graph fusion, speculative decoding • Real-world benchmarks: 2–3x speedup on T5 models with minimal accuracy loss • Deployment strategies: sharding, OTA updates, version control for safe rollouts • Platform-specific tuning: custom Metal kernel optimizations for Apple silicon If you’re building for real-world, low-latency LLM deployment, this is worth a watch. 💡 Some eye-opening takeaways: • INT4 quantization + speculative decoding pushed T5-Large from 58 → 140 tokens/sec, and T5-Base to 300 tokens/sec • LLaMA 3.2B went from 68 → 210 tokens/sec with smart quantization • Grammarly keeps their keyboard model under 66MB and deploys via OTA updates with rollback safeguards Here's the full video: https://lnkd.in/dhk8jRFi

6